
Loading...

Loading...
A short guide to running a LightChain AI worker with the app. Everything here is handled for you by one-click install and the dashboard. This is just so you know what is happening under the hood.
A worker is your machine serving AI inference jobs for the LightChain network and earning LCAI. It runs a local model through Ollama inside a Docker container, and it is identified on-chain by a generated worker key with a staked deposit. The app generates that key, funds and stakes it from a wallet you choose, starts the container, and keeps it online.
Set up: pick your model(s), set a keystore password, fund the generated worker address, and the app registers and starts the worker.
Earn: the gateway routes jobs to you automatically. Each completed job's reward is escrowed, then released by the network (about hourly, up to a few hours).
Settle and withdraw: Settle earnings moves released rewards into your worker wallet; Withdraw Funds sends that balance to any address you choose.
Exit: Deregister returns your stake to the worker wallet and stops the worker. Then withdraw to take everything out. If a job your worker accepted but never finished is blocking the exit, Clear stuck jobs times it out first (on mainnet that realizes a small slash).
Your stake is locked in the registry while you are registered. Released earnings sit in the worker wallet as spendable LCAI. The stake only returns to the wallet when you deregister, so a typical end-of-life balance is earnings plus the returned stake, which you then withdraw.
You get slashed for going silent on a job you accepted (acknowledged, then failed to finish in time), not for honest, reported failures. The two defenses are a worker that stays online and a model that stays warm. The app installs a keep-online watchdog (auto-starts Docker and the worker, keeps your machine awake while it runs) and pins the model in memory so the first job never pays a cold load. The Speed test shows your worst-case job time against the deadline before it matters.
A worker can serve a single model or several at once. Every served model has to stay loaded in memory at the same time, so the model picker sums their footprints and warns when a set will not fit your machine. Each model has its own fee. Change the set live on the dashboard with Models this worker serves; it updates the set on-chain with no re-stake and restarts the worker with it.
Your worker key and keystore password are generated and kept on your device only, in the OS keychain and an encrypted keystore. They are never sent to any server, and all signing happens locally. If you ever replace a key, the old one is archived on your device, so a worker that still holds a stake is never lost. Recover a replaced key on the dashboard lists them, flags any still staked, and restores one as your active worker.
The app wraps the official LightChain worker toolkit. If you would rather run every step yourself, the toolkit is the upstream source.